It depends, on a couple of factors: the transcriptome size, degree of annotations etc- I would not go below 25M.
As a very general rule, I would say a bare minimum of 10M reads per sample should be targeted, but, you will get much more reliable and robust results with more (25M-50M reads per sample would be great).
Do you have a reference genome for wheat? I think yes but If you do not, then your sequencing needs will be much, much greater than if you had one, as you will likely want to do your own de novo assembly to map your reads back to in order to analyze differential expression. If you need to build your own reference, then you will need much greater coverage than if you could map reads directly to an established reference genome.
This article should be helpfull for you:Article RNA-seq differential expression studies: More sequence or mo...
I think one best way is to ask the companies which have done transcriptome sequencing in wheat. They will provide any information you want to know. By the way, you will get a quality control report after RNA seq. It will show that as the reads increase, the identification of new transcripts will reach a plateau.
We were also asked companies and they said 30M reads enough. but some researchers reported less than 50M reads not enough. So I looking for some paper about this. Thanks for answer Chuan Liu
First, you have to specify what's the goal of the experiment - is it differential gene expression, transcript expression, assembly, or something completely different? Very simply put - the depth (length is something different) of RNA-Seq depends on a) the goal of the experiment (see above), b) the target expression of genes/transcripts you are interested in (high expressed genes will need less depth than low expressed genes), c) size of the transcriptome (not genome, as Marta Jaskulak mentioned) and gene sizes; you sequence expressed genes, not genome itself. Don't trust the companies telling you 30M is enough. I don't think they have a lot of experience with wheat or plants in general. 30M is generally recommended for human RNA-Seq targeting expression of high expressed protein-coding genes. Wheat (if we talk about T. aestivum) has much bigger transcriptome (http://plants.ensembl.org/Triticum_aestivum/Info/Annotation/#genebuild). Human has ~21k expressed protein-coding genes. Wheat has almost 5x as many! I don't know about the length nor about the expression of wheat genes but I would assume that you will find much more expressed genes in wheat than in human.
If you have time, I would recommend you to simulate wheat gene expression and draw a conclusion from there. You can also check already existing publications with wheat RNA-Seq, see how many reads they sequenced and perform a 'mock' analysis to see how the expression behaves depending on the number of reads (make several downsamplings, calculated DE if it's your goal, compare results for genes at different expression levels).